image
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill recommends using official APIs and well-known services for image generation, such as Google Gemini, OpenAI, and Black Forest Labs. All referenced URLs (e.g., ai.google.dev, docs.bfl.ai) are official service domains.
- [SAFE]: The skill gathers project-specific context from local files like .agents/product-marketing.md to improve the relevance of its marketing advice.
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting data from local context files and suggesting the use of curl to fetch content from external URLs. These ingestion points lack explicit sanitization or boundary markers. Evidence: 1. Ingestion points: Local files (.agents/product-marketing.md) and curl output from user-specified domains. 2. Boundary markers: None. 3. Capability inventory: Shell command execution and file system access. 4. Sanitization: None.
- [COMMAND_EXECUTION]: The skill suggests several CLI tools for image optimization and analysis, including cwebp, mogrify, jpegoptim, and curl. These are standard developer tools and do not represent a malicious pattern.
Audit Metadata